e2e

E2E tests for the operator

E2E tests run on a Kind cluster that is set up by the make target. In the case of an error, the cluster stays intact, and can be removed by calling: kind delete clusters intel-gpu-base-operator-test-e2e

The node where the e2e tests are run has to have one or more Intel GPUs with suitable KMD installed. While the operator prioritizes the newer xe KMD, it also supports the i915.

Dependencies

Runtime dependencies:

• make - to trigger e2e run
• go - to compile and run the e2e tests
• kubectl - to deploy the operator and overall access to the cluster
• helm - to deploy the operator
• kind - to setup a temporary cluster
• jq - to parse json data in tests

Helm and Kind can be installed using make:

mkdir -p $(pwd)/bin
export PATH=$PATH:$(pwd)/bin
make helm
make kind

Collateral images

Test run automatically pre-pulls the images used in the run to prevent timeouts during the test execution.

  STEP: loading the manager(Operator) image on Kind @ 12/03/25 10:33:33.238
  STEP: loading collateral images on Kind @ 12/03/25 10:33:33.349
  STEP: loading image intel/intel-gpu-plugin:0.34.0 to kind cluster @ 12/03/25 10:33:33.35
  STEP: loading image intel/intel-gpu-levelzero:0.34.0 to kind cluster @ 12/03/25 10:33:35.087
  STEP: loading image intel/xpumanager:v1.2.27 to kind cluster @ 12/03/25 10:33:36.941
  STEP: loading image ghcr.io/intel/intel-resource-drivers-for-kubernetes/intel-gpu-resource-driver:v0.9.0 to kind cluster @ 12/03/25 10:33:38.787

How to run

Build the latest operator:

make docker-build

Test run automatically imports the operator image to Kind from docker.

And run the tests for nodes with newer xe KMD (i915 tests are excluded):

make test-e2e

For nodes with older i915 KMD (xe tests are excluded):

make test-e2e-i915

The separation of xe and i915 is due to how the Intel GPU plugin registers two different monitoring resources. The XPU Manager has to have a Daemonset targeted for either resource.